What's recommended for the number of users to use in qualitative and quantitative usability studies?

Question

It was recommended to me on the User Interface Stack Exchange that I also ask this question here. So....

I'm currently rebuilding our entire intranet from scratch, mostly because the tech behind is out-dated and it has been proved that a lot of information is difficult to find.

Though that is beside the point, what I am wondering is what would be the optimum amount of users to use for qualitative and quantitative testing with a userbase of around 1000 users?

Is there a general rule of thumb for both based on the total number of users you have? or is it just say 5 for qualitative and 10 for quantitative?

What would be the best approach?

Answer 1

Not an exact answer to your question (which is kind of not understandable enough for conrete answers, see Srikant's comment), but a more general one:
If you want to do statistical testing on your data it is advisable to make sure that your sample conforms to the central limit theorem. That is, the distribution of sample means from which your sample is drawn is normal (i.e., bell-shaped). Then you can use a lot of nice methods (as long as the other assumptions are met) on your data that try to estimate population parameters from your sample. A rule of thumb about the central limit theorem is that you need at least 30 participants (or 20, see here).
My advice: If you think that 10 could be enough, use at least 20.

For qualitative testing, I would guess, just use participants until you get a picture of what the issues are. Ask a couple of people to test your stuff and ask them to tell you their impression. If they agree to some degree, there you go. I think 5 is a good starting point (caveat: this is a non-theoretically answer, based on my stomach, and I just had lunch).

If you are interested in polls from your sample, this discussion could be useful.

Answer 2

Jakob Nielsen recommends testing with five users for optimal results. This assertion has been challenged a few times, both empirically and theoretically, but generally seems to hold quite well.